A Statistical Model for the Influence of Temperature on Bike Demand in Bike-sharing Systems

Abstract

Efficient fleet management is essential for bike-sharing systems. Thus, it is important to understand the impact of environmental factors on bike demand. This thesis proposes a method to analyze the influence of temperature on bike demand. Hourly temperature data are approximated by smoothed curves and modeled by functional principal components. Bike check-out times, which can be seen as realizations of a doubly stochastic process, are modeled using multiplicative component models on the underlying intensity functions. The respective component scores are then related via a multivariate regression model. An analysis of data from the Divvy system of the City of Chicago is presented as an example of application.